xref: /linux/fs/ceph/super.c (revision 0889d07f3e4b171c453b2aaf2b257f9074cdf624)
1 // SPDX-License-Identifier: GPL-2.0-only
2 
3 #include <linux/ceph/ceph_debug.h>
4 
5 #include <linux/backing-dev.h>
6 #include <linux/ctype.h>
7 #include <linux/fs.h>
8 #include <linux/inet.h>
9 #include <linux/in6.h>
10 #include <linux/module.h>
11 #include <linux/mount.h>
12 #include <linux/parser.h>
13 #include <linux/sched.h>
14 #include <linux/seq_file.h>
15 #include <linux/slab.h>
16 #include <linux/statfs.h>
17 #include <linux/string.h>
18 
19 #include "super.h"
20 #include "mds_client.h"
21 #include "cache.h"
22 
23 #include <linux/ceph/ceph_features.h>
24 #include <linux/ceph/decode.h>
25 #include <linux/ceph/mon_client.h>
26 #include <linux/ceph/auth.h>
27 #include <linux/ceph/debugfs.h>
28 
29 /*
30  * Ceph superblock operations
31  *
32  * Handle the basics of mounting, unmounting.
33  */
34 
35 /*
36  * super ops
37  */
38 static void ceph_put_super(struct super_block *s)
39 {
40 	struct ceph_fs_client *fsc = ceph_sb_to_client(s);
41 
42 	dout("put_super\n");
43 	ceph_mdsc_close_sessions(fsc->mdsc);
44 }
45 
46 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
47 {
48 	struct ceph_fs_client *fsc = ceph_inode_to_client(d_inode(dentry));
49 	struct ceph_mon_client *monc = &fsc->client->monc;
50 	struct ceph_statfs st;
51 	u64 fsid;
52 	int err;
53 	u64 data_pool;
54 
55 	if (fsc->mdsc->mdsmap->m_num_data_pg_pools == 1) {
56 		data_pool = fsc->mdsc->mdsmap->m_data_pg_pools[0];
57 	} else {
58 		data_pool = CEPH_NOPOOL;
59 	}
60 
61 	dout("statfs\n");
62 	err = ceph_monc_do_statfs(monc, data_pool, &st);
63 	if (err < 0)
64 		return err;
65 
66 	/* fill in kstatfs */
67 	buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
68 
69 	/*
70 	 * express utilization in terms of large blocks to avoid
71 	 * overflow on 32-bit machines.
72 	 *
73 	 * NOTE: for the time being, we make bsize == frsize to humor
74 	 * not-yet-ancient versions of glibc that are broken.
75 	 * Someday, we will probably want to report a real block
76 	 * size...  whatever that may mean for a network file system!
77 	 */
78 	buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
79 	buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
80 
81 	/*
82 	 * By default use root quota for stats; fallback to overall filesystem
83 	 * usage if using 'noquotadf' mount option or if the root dir doesn't
84 	 * have max_bytes quota set.
85 	 */
86 	if (ceph_test_mount_opt(fsc, NOQUOTADF) ||
87 	    !ceph_quota_update_statfs(fsc, buf)) {
88 		buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
89 		buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
90 		buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
91 	}
92 
93 	buf->f_files = le64_to_cpu(st.num_objects);
94 	buf->f_ffree = -1;
95 	buf->f_namelen = NAME_MAX;
96 
97 	/* Must convert the fsid, for consistent values across arches */
98 	mutex_lock(&monc->mutex);
99 	fsid = le64_to_cpu(*(__le64 *)(&monc->monmap->fsid)) ^
100 	       le64_to_cpu(*((__le64 *)&monc->monmap->fsid + 1));
101 	mutex_unlock(&monc->mutex);
102 
103 	buf->f_fsid.val[0] = fsid & 0xffffffff;
104 	buf->f_fsid.val[1] = fsid >> 32;
105 
106 	return 0;
107 }
108 
109 
110 static int ceph_sync_fs(struct super_block *sb, int wait)
111 {
112 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
113 
114 	if (!wait) {
115 		dout("sync_fs (non-blocking)\n");
116 		ceph_flush_dirty_caps(fsc->mdsc);
117 		dout("sync_fs (non-blocking) done\n");
118 		return 0;
119 	}
120 
121 	dout("sync_fs (blocking)\n");
122 	ceph_osdc_sync(&fsc->client->osdc);
123 	ceph_mdsc_sync(fsc->mdsc);
124 	dout("sync_fs (blocking) done\n");
125 	return 0;
126 }
127 
128 /*
129  * mount options
130  */
131 enum {
132 	Opt_wsize,
133 	Opt_rsize,
134 	Opt_rasize,
135 	Opt_caps_wanted_delay_min,
136 	Opt_caps_wanted_delay_max,
137 	Opt_caps_max,
138 	Opt_readdir_max_entries,
139 	Opt_readdir_max_bytes,
140 	Opt_congestion_kb,
141 	Opt_last_int,
142 	/* int args above */
143 	Opt_snapdirname,
144 	Opt_mds_namespace,
145 	Opt_fscache_uniq,
146 	Opt_recover_session,
147 	Opt_last_string,
148 	/* string args above */
149 	Opt_dirstat,
150 	Opt_nodirstat,
151 	Opt_rbytes,
152 	Opt_norbytes,
153 	Opt_asyncreaddir,
154 	Opt_noasyncreaddir,
155 	Opt_dcache,
156 	Opt_nodcache,
157 	Opt_ino32,
158 	Opt_noino32,
159 	Opt_fscache,
160 	Opt_nofscache,
161 	Opt_poolperm,
162 	Opt_nopoolperm,
163 	Opt_require_active_mds,
164 	Opt_norequire_active_mds,
165 #ifdef CONFIG_CEPH_FS_POSIX_ACL
166 	Opt_acl,
167 #endif
168 	Opt_noacl,
169 	Opt_quotadf,
170 	Opt_noquotadf,
171 	Opt_copyfrom,
172 	Opt_nocopyfrom,
173 };
174 
175 static match_table_t fsopt_tokens = {
176 	{Opt_wsize, "wsize=%d"},
177 	{Opt_rsize, "rsize=%d"},
178 	{Opt_rasize, "rasize=%d"},
179 	{Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
180 	{Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
181 	{Opt_caps_max, "caps_max=%d"},
182 	{Opt_readdir_max_entries, "readdir_max_entries=%d"},
183 	{Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
184 	{Opt_congestion_kb, "write_congestion_kb=%d"},
185 	/* int args above */
186 	{Opt_snapdirname, "snapdirname=%s"},
187 	{Opt_mds_namespace, "mds_namespace=%s"},
188 	{Opt_recover_session, "recover_session=%s"},
189 	{Opt_fscache_uniq, "fsc=%s"},
190 	/* string args above */
191 	{Opt_dirstat, "dirstat"},
192 	{Opt_nodirstat, "nodirstat"},
193 	{Opt_rbytes, "rbytes"},
194 	{Opt_norbytes, "norbytes"},
195 	{Opt_asyncreaddir, "asyncreaddir"},
196 	{Opt_noasyncreaddir, "noasyncreaddir"},
197 	{Opt_dcache, "dcache"},
198 	{Opt_nodcache, "nodcache"},
199 	{Opt_ino32, "ino32"},
200 	{Opt_noino32, "noino32"},
201 	{Opt_fscache, "fsc"},
202 	{Opt_nofscache, "nofsc"},
203 	{Opt_poolperm, "poolperm"},
204 	{Opt_nopoolperm, "nopoolperm"},
205 	{Opt_require_active_mds, "require_active_mds"},
206 	{Opt_norequire_active_mds, "norequire_active_mds"},
207 #ifdef CONFIG_CEPH_FS_POSIX_ACL
208 	{Opt_acl, "acl"},
209 #endif
210 	{Opt_noacl, "noacl"},
211 	{Opt_quotadf, "quotadf"},
212 	{Opt_noquotadf, "noquotadf"},
213 	{Opt_copyfrom, "copyfrom"},
214 	{Opt_nocopyfrom, "nocopyfrom"},
215 	{-1, NULL}
216 };
217 
218 static int parse_fsopt_token(char *c, void *private)
219 {
220 	struct ceph_mount_options *fsopt = private;
221 	substring_t argstr[MAX_OPT_ARGS];
222 	int token, intval, ret;
223 
224 	token = match_token((char *)c, fsopt_tokens, argstr);
225 	if (token < 0)
226 		return -EINVAL;
227 
228 	if (token < Opt_last_int) {
229 		ret = match_int(&argstr[0], &intval);
230 		if (ret < 0) {
231 			pr_err("bad option arg (not int) at '%s'\n", c);
232 			return ret;
233 		}
234 		dout("got int token %d val %d\n", token, intval);
235 	} else if (token > Opt_last_int && token < Opt_last_string) {
236 		dout("got string token %d val %s\n", token,
237 		     argstr[0].from);
238 	} else {
239 		dout("got token %d\n", token);
240 	}
241 
242 	switch (token) {
243 	case Opt_snapdirname:
244 		kfree(fsopt->snapdir_name);
245 		fsopt->snapdir_name = kstrndup(argstr[0].from,
246 					       argstr[0].to-argstr[0].from,
247 					       GFP_KERNEL);
248 		if (!fsopt->snapdir_name)
249 			return -ENOMEM;
250 		break;
251 	case Opt_mds_namespace:
252 		kfree(fsopt->mds_namespace);
253 		fsopt->mds_namespace = kstrndup(argstr[0].from,
254 						argstr[0].to-argstr[0].from,
255 						GFP_KERNEL);
256 		if (!fsopt->mds_namespace)
257 			return -ENOMEM;
258 		break;
259 	case Opt_recover_session:
260 		if (!strncmp(argstr[0].from, "no",
261 			     argstr[0].to - argstr[0].from)) {
262 			fsopt->flags &= ~CEPH_MOUNT_OPT_CLEANRECOVER;
263 		} else if (!strncmp(argstr[0].from, "clean",
264 				    argstr[0].to - argstr[0].from)) {
265 			fsopt->flags |= CEPH_MOUNT_OPT_CLEANRECOVER;
266 		} else {
267 			return -EINVAL;
268 		}
269 		break;
270 	case Opt_fscache_uniq:
271 		kfree(fsopt->fscache_uniq);
272 		fsopt->fscache_uniq = kstrndup(argstr[0].from,
273 					       argstr[0].to-argstr[0].from,
274 					       GFP_KERNEL);
275 		if (!fsopt->fscache_uniq)
276 			return -ENOMEM;
277 		fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
278 		break;
279 		/* misc */
280 	case Opt_wsize:
281 		if (intval < (int)PAGE_SIZE || intval > CEPH_MAX_WRITE_SIZE)
282 			return -EINVAL;
283 		fsopt->wsize = ALIGN(intval, PAGE_SIZE);
284 		break;
285 	case Opt_rsize:
286 		if (intval < (int)PAGE_SIZE || intval > CEPH_MAX_READ_SIZE)
287 			return -EINVAL;
288 		fsopt->rsize = ALIGN(intval, PAGE_SIZE);
289 		break;
290 	case Opt_rasize:
291 		if (intval < 0)
292 			return -EINVAL;
293 		fsopt->rasize = ALIGN(intval, PAGE_SIZE);
294 		break;
295 	case Opt_caps_wanted_delay_min:
296 		if (intval < 1)
297 			return -EINVAL;
298 		fsopt->caps_wanted_delay_min = intval;
299 		break;
300 	case Opt_caps_wanted_delay_max:
301 		if (intval < 1)
302 			return -EINVAL;
303 		fsopt->caps_wanted_delay_max = intval;
304 		break;
305 	case Opt_caps_max:
306 		if (intval < 0)
307 			return -EINVAL;
308 		fsopt->caps_max = intval;
309 		break;
310 	case Opt_readdir_max_entries:
311 		if (intval < 1)
312 			return -EINVAL;
313 		fsopt->max_readdir = intval;
314 		break;
315 	case Opt_readdir_max_bytes:
316 		if (intval < (int)PAGE_SIZE && intval != 0)
317 			return -EINVAL;
318 		fsopt->max_readdir_bytes = intval;
319 		break;
320 	case Opt_congestion_kb:
321 		if (intval < 1024) /* at least 1M */
322 			return -EINVAL;
323 		fsopt->congestion_kb = intval;
324 		break;
325 	case Opt_dirstat:
326 		fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
327 		break;
328 	case Opt_nodirstat:
329 		fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
330 		break;
331 	case Opt_rbytes:
332 		fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
333 		break;
334 	case Opt_norbytes:
335 		fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
336 		break;
337 	case Opt_asyncreaddir:
338 		fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
339 		break;
340 	case Opt_noasyncreaddir:
341 		fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
342 		break;
343 	case Opt_dcache:
344 		fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
345 		break;
346 	case Opt_nodcache:
347 		fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
348 		break;
349 	case Opt_ino32:
350 		fsopt->flags |= CEPH_MOUNT_OPT_INO32;
351 		break;
352 	case Opt_noino32:
353 		fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
354 		break;
355 	case Opt_fscache:
356 		fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
357 		kfree(fsopt->fscache_uniq);
358 		fsopt->fscache_uniq = NULL;
359 		break;
360 	case Opt_nofscache:
361 		fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
362 		kfree(fsopt->fscache_uniq);
363 		fsopt->fscache_uniq = NULL;
364 		break;
365 	case Opt_poolperm:
366 		fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
367 		break;
368 	case Opt_nopoolperm:
369 		fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
370 		break;
371 	case Opt_require_active_mds:
372 		fsopt->flags &= ~CEPH_MOUNT_OPT_MOUNTWAIT;
373 		break;
374 	case Opt_norequire_active_mds:
375 		fsopt->flags |= CEPH_MOUNT_OPT_MOUNTWAIT;
376 		break;
377 	case Opt_quotadf:
378 		fsopt->flags &= ~CEPH_MOUNT_OPT_NOQUOTADF;
379 		break;
380 	case Opt_noquotadf:
381 		fsopt->flags |= CEPH_MOUNT_OPT_NOQUOTADF;
382 		break;
383 	case Opt_copyfrom:
384 		fsopt->flags &= ~CEPH_MOUNT_OPT_NOCOPYFROM;
385 		break;
386 	case Opt_nocopyfrom:
387 		fsopt->flags |= CEPH_MOUNT_OPT_NOCOPYFROM;
388 		break;
389 #ifdef CONFIG_CEPH_FS_POSIX_ACL
390 	case Opt_acl:
391 		fsopt->sb_flags |= SB_POSIXACL;
392 		break;
393 #endif
394 	case Opt_noacl:
395 		fsopt->sb_flags &= ~SB_POSIXACL;
396 		break;
397 	default:
398 		BUG_ON(token);
399 	}
400 	return 0;
401 }
402 
403 static void destroy_mount_options(struct ceph_mount_options *args)
404 {
405 	dout("destroy_mount_options %p\n", args);
406 	kfree(args->snapdir_name);
407 	kfree(args->mds_namespace);
408 	kfree(args->server_path);
409 	kfree(args->fscache_uniq);
410 	kfree(args);
411 }
412 
413 static int strcmp_null(const char *s1, const char *s2)
414 {
415 	if (!s1 && !s2)
416 		return 0;
417 	if (s1 && !s2)
418 		return -1;
419 	if (!s1 && s2)
420 		return 1;
421 	return strcmp(s1, s2);
422 }
423 
424 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
425 				 struct ceph_options *new_opt,
426 				 struct ceph_fs_client *fsc)
427 {
428 	struct ceph_mount_options *fsopt1 = new_fsopt;
429 	struct ceph_mount_options *fsopt2 = fsc->mount_options;
430 	int ofs = offsetof(struct ceph_mount_options, snapdir_name);
431 	int ret;
432 
433 	ret = memcmp(fsopt1, fsopt2, ofs);
434 	if (ret)
435 		return ret;
436 
437 	ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
438 	if (ret)
439 		return ret;
440 	ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace);
441 	if (ret)
442 		return ret;
443 	ret = strcmp_null(fsopt1->server_path, fsopt2->server_path);
444 	if (ret)
445 		return ret;
446 	ret = strcmp_null(fsopt1->fscache_uniq, fsopt2->fscache_uniq);
447 	if (ret)
448 		return ret;
449 
450 	return ceph_compare_options(new_opt, fsc->client);
451 }
452 
453 static int parse_mount_options(struct ceph_mount_options **pfsopt,
454 			       struct ceph_options **popt,
455 			       int flags, char *options,
456 			       const char *dev_name)
457 {
458 	struct ceph_mount_options *fsopt;
459 	const char *dev_name_end;
460 	int err;
461 
462 	if (!dev_name || !*dev_name)
463 		return -EINVAL;
464 
465 	fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
466 	if (!fsopt)
467 		return -ENOMEM;
468 
469 	dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
470 
471 	fsopt->sb_flags = flags;
472 	fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
473 
474 	fsopt->wsize = CEPH_MAX_WRITE_SIZE;
475 	fsopt->rsize = CEPH_MAX_READ_SIZE;
476 	fsopt->rasize = CEPH_RASIZE_DEFAULT;
477 	fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
478 	if (!fsopt->snapdir_name) {
479 		err = -ENOMEM;
480 		goto out;
481 	}
482 
483 	fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
484 	fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
485 	fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
486 	fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
487 	fsopt->congestion_kb = default_congestion_kb();
488 
489 	/*
490 	 * Distinguish the server list from the path in "dev_name".
491 	 * Internally we do not include the leading '/' in the path.
492 	 *
493 	 * "dev_name" will look like:
494 	 *     <server_spec>[,<server_spec>...]:[<path>]
495 	 * where
496 	 *     <server_spec> is <ip>[:<port>]
497 	 *     <path> is optional, but if present must begin with '/'
498 	 */
499 	dev_name_end = strchr(dev_name, '/');
500 	if (dev_name_end) {
501 		if (strlen(dev_name_end) > 1) {
502 			fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
503 			if (!fsopt->server_path) {
504 				err = -ENOMEM;
505 				goto out;
506 			}
507 		}
508 	} else {
509 		dev_name_end = dev_name + strlen(dev_name);
510 	}
511 	err = -EINVAL;
512 	dev_name_end--;		/* back up to ':' separator */
513 	if (dev_name_end < dev_name || *dev_name_end != ':') {
514 		pr_err("device name is missing path (no : separator in %s)\n",
515 				dev_name);
516 		goto out;
517 	}
518 	dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
519 	if (fsopt->server_path)
520 		dout("server path '%s'\n", fsopt->server_path);
521 
522 	*popt = ceph_parse_options(options, dev_name, dev_name_end,
523 				 parse_fsopt_token, (void *)fsopt);
524 	if (IS_ERR(*popt)) {
525 		err = PTR_ERR(*popt);
526 		goto out;
527 	}
528 
529 	/* success */
530 	*pfsopt = fsopt;
531 	return 0;
532 
533 out:
534 	destroy_mount_options(fsopt);
535 	return err;
536 }
537 
538 /**
539  * ceph_show_options - Show mount options in /proc/mounts
540  * @m: seq_file to write to
541  * @root: root of that (sub)tree
542  */
543 static int ceph_show_options(struct seq_file *m, struct dentry *root)
544 {
545 	struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
546 	struct ceph_mount_options *fsopt = fsc->mount_options;
547 	size_t pos;
548 	int ret;
549 
550 	/* a comma between MNT/MS and client options */
551 	seq_putc(m, ',');
552 	pos = m->count;
553 
554 	ret = ceph_print_client_options(m, fsc->client, false);
555 	if (ret)
556 		return ret;
557 
558 	/* retract our comma if no client options */
559 	if (m->count == pos)
560 		m->count--;
561 
562 	if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
563 		seq_puts(m, ",dirstat");
564 	if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
565 		seq_puts(m, ",rbytes");
566 	if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
567 		seq_puts(m, ",noasyncreaddir");
568 	if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
569 		seq_puts(m, ",nodcache");
570 	if (fsopt->flags & CEPH_MOUNT_OPT_INO32)
571 		seq_puts(m, ",ino32");
572 	if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) {
573 		seq_show_option(m, "fsc", fsopt->fscache_uniq);
574 	}
575 	if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
576 		seq_puts(m, ",nopoolperm");
577 	if (fsopt->flags & CEPH_MOUNT_OPT_NOQUOTADF)
578 		seq_puts(m, ",noquotadf");
579 
580 #ifdef CONFIG_CEPH_FS_POSIX_ACL
581 	if (fsopt->sb_flags & SB_POSIXACL)
582 		seq_puts(m, ",acl");
583 	else
584 		seq_puts(m, ",noacl");
585 #endif
586 
587 	if ((fsopt->flags & CEPH_MOUNT_OPT_NOCOPYFROM) == 0)
588 		seq_puts(m, ",copyfrom");
589 
590 	if (fsopt->mds_namespace)
591 		seq_show_option(m, "mds_namespace", fsopt->mds_namespace);
592 
593 	if (fsopt->flags & CEPH_MOUNT_OPT_CLEANRECOVER)
594 		seq_show_option(m, "recover_session", "clean");
595 
596 	if (fsopt->wsize != CEPH_MAX_WRITE_SIZE)
597 		seq_printf(m, ",wsize=%d", fsopt->wsize);
598 	if (fsopt->rsize != CEPH_MAX_READ_SIZE)
599 		seq_printf(m, ",rsize=%d", fsopt->rsize);
600 	if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
601 		seq_printf(m, ",rasize=%d", fsopt->rasize);
602 	if (fsopt->congestion_kb != default_congestion_kb())
603 		seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
604 	if (fsopt->caps_max)
605 		seq_printf(m, ",caps_max=%d", fsopt->caps_max);
606 	if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
607 		seq_printf(m, ",caps_wanted_delay_min=%d",
608 			 fsopt->caps_wanted_delay_min);
609 	if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
610 		seq_printf(m, ",caps_wanted_delay_max=%d",
611 			   fsopt->caps_wanted_delay_max);
612 	if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
613 		seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
614 	if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
615 		seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
616 	if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
617 		seq_show_option(m, "snapdirname", fsopt->snapdir_name);
618 
619 	return 0;
620 }
621 
622 /*
623  * handle any mon messages the standard library doesn't understand.
624  * return error if we don't either.
625  */
626 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
627 {
628 	struct ceph_fs_client *fsc = client->private;
629 	int type = le16_to_cpu(msg->hdr.type);
630 
631 	switch (type) {
632 	case CEPH_MSG_MDS_MAP:
633 		ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
634 		return 0;
635 	case CEPH_MSG_FS_MAP_USER:
636 		ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
637 		return 0;
638 	default:
639 		return -1;
640 	}
641 }
642 
643 /*
644  * create a new fs client
645  *
646  * Success or not, this function consumes @fsopt and @opt.
647  */
648 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
649 					struct ceph_options *opt)
650 {
651 	struct ceph_fs_client *fsc;
652 	int page_count;
653 	size_t size;
654 	int err;
655 
656 	fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
657 	if (!fsc) {
658 		err = -ENOMEM;
659 		goto fail;
660 	}
661 
662 	fsc->client = ceph_create_client(opt, fsc);
663 	if (IS_ERR(fsc->client)) {
664 		err = PTR_ERR(fsc->client);
665 		goto fail;
666 	}
667 	opt = NULL; /* fsc->client now owns this */
668 
669 	fsc->client->extra_mon_dispatch = extra_mon_dispatch;
670 	ceph_set_opt(fsc->client, ABORT_ON_FULL);
671 
672 	if (!fsopt->mds_namespace) {
673 		ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
674 				   0, true);
675 	} else {
676 		ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
677 				   0, false);
678 	}
679 
680 	fsc->mount_options = fsopt;
681 
682 	fsc->sb = NULL;
683 	fsc->mount_state = CEPH_MOUNT_MOUNTING;
684 	fsc->filp_gen = 1;
685 
686 	atomic_long_set(&fsc->writeback_count, 0);
687 
688 	err = -ENOMEM;
689 	/*
690 	 * The number of concurrent works can be high but they don't need
691 	 * to be processed in parallel, limit concurrency.
692 	 */
693 	fsc->inode_wq = alloc_workqueue("ceph-inode", WQ_UNBOUND, 0);
694 	if (!fsc->inode_wq)
695 		goto fail_client;
696 	fsc->cap_wq = alloc_workqueue("ceph-cap", 0, 1);
697 	if (!fsc->cap_wq)
698 		goto fail_inode_wq;
699 
700 	/* set up mempools */
701 	err = -ENOMEM;
702 	page_count = fsc->mount_options->wsize >> PAGE_SHIFT;
703 	size = sizeof (struct page *) * (page_count ? page_count : 1);
704 	fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10, size);
705 	if (!fsc->wb_pagevec_pool)
706 		goto fail_cap_wq;
707 
708 	return fsc;
709 
710 fail_cap_wq:
711 	destroy_workqueue(fsc->cap_wq);
712 fail_inode_wq:
713 	destroy_workqueue(fsc->inode_wq);
714 fail_client:
715 	ceph_destroy_client(fsc->client);
716 fail:
717 	kfree(fsc);
718 	if (opt)
719 		ceph_destroy_options(opt);
720 	destroy_mount_options(fsopt);
721 	return ERR_PTR(err);
722 }
723 
724 static void flush_fs_workqueues(struct ceph_fs_client *fsc)
725 {
726 	flush_workqueue(fsc->inode_wq);
727 	flush_workqueue(fsc->cap_wq);
728 }
729 
730 static void destroy_fs_client(struct ceph_fs_client *fsc)
731 {
732 	dout("destroy_fs_client %p\n", fsc);
733 
734 	ceph_mdsc_destroy(fsc);
735 	destroy_workqueue(fsc->inode_wq);
736 	destroy_workqueue(fsc->cap_wq);
737 
738 	mempool_destroy(fsc->wb_pagevec_pool);
739 
740 	destroy_mount_options(fsc->mount_options);
741 
742 	ceph_destroy_client(fsc->client);
743 
744 	kfree(fsc);
745 	dout("destroy_fs_client %p done\n", fsc);
746 }
747 
748 /*
749  * caches
750  */
751 struct kmem_cache *ceph_inode_cachep;
752 struct kmem_cache *ceph_cap_cachep;
753 struct kmem_cache *ceph_cap_flush_cachep;
754 struct kmem_cache *ceph_dentry_cachep;
755 struct kmem_cache *ceph_file_cachep;
756 struct kmem_cache *ceph_dir_file_cachep;
757 
758 static void ceph_inode_init_once(void *foo)
759 {
760 	struct ceph_inode_info *ci = foo;
761 	inode_init_once(&ci->vfs_inode);
762 }
763 
764 static int __init init_caches(void)
765 {
766 	int error = -ENOMEM;
767 
768 	ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
769 				      sizeof(struct ceph_inode_info),
770 				      __alignof__(struct ceph_inode_info),
771 				      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
772 				      SLAB_ACCOUNT, ceph_inode_init_once);
773 	if (!ceph_inode_cachep)
774 		return -ENOMEM;
775 
776 	ceph_cap_cachep = KMEM_CACHE(ceph_cap, SLAB_MEM_SPREAD);
777 	if (!ceph_cap_cachep)
778 		goto bad_cap;
779 	ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
780 					   SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
781 	if (!ceph_cap_flush_cachep)
782 		goto bad_cap_flush;
783 
784 	ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
785 					SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
786 	if (!ceph_dentry_cachep)
787 		goto bad_dentry;
788 
789 	ceph_file_cachep = KMEM_CACHE(ceph_file_info, SLAB_MEM_SPREAD);
790 	if (!ceph_file_cachep)
791 		goto bad_file;
792 
793 	ceph_dir_file_cachep = KMEM_CACHE(ceph_dir_file_info, SLAB_MEM_SPREAD);
794 	if (!ceph_dir_file_cachep)
795 		goto bad_dir_file;
796 
797 	error = ceph_fscache_register();
798 	if (error)
799 		goto bad_fscache;
800 
801 	return 0;
802 
803 bad_fscache:
804 	kmem_cache_destroy(ceph_dir_file_cachep);
805 bad_dir_file:
806 	kmem_cache_destroy(ceph_file_cachep);
807 bad_file:
808 	kmem_cache_destroy(ceph_dentry_cachep);
809 bad_dentry:
810 	kmem_cache_destroy(ceph_cap_flush_cachep);
811 bad_cap_flush:
812 	kmem_cache_destroy(ceph_cap_cachep);
813 bad_cap:
814 	kmem_cache_destroy(ceph_inode_cachep);
815 	return error;
816 }
817 
818 static void destroy_caches(void)
819 {
820 	/*
821 	 * Make sure all delayed rcu free inodes are flushed before we
822 	 * destroy cache.
823 	 */
824 	rcu_barrier();
825 
826 	kmem_cache_destroy(ceph_inode_cachep);
827 	kmem_cache_destroy(ceph_cap_cachep);
828 	kmem_cache_destroy(ceph_cap_flush_cachep);
829 	kmem_cache_destroy(ceph_dentry_cachep);
830 	kmem_cache_destroy(ceph_file_cachep);
831 	kmem_cache_destroy(ceph_dir_file_cachep);
832 
833 	ceph_fscache_unregister();
834 }
835 
836 
837 /*
838  * ceph_umount_begin - initiate forced umount.  Tear down down the
839  * mount, skipping steps that may hang while waiting for server(s).
840  */
841 static void ceph_umount_begin(struct super_block *sb)
842 {
843 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
844 
845 	dout("ceph_umount_begin - starting forced umount\n");
846 	if (!fsc)
847 		return;
848 	fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
849 	ceph_osdc_abort_requests(&fsc->client->osdc, -EIO);
850 	ceph_mdsc_force_umount(fsc->mdsc);
851 	fsc->filp_gen++; // invalidate open files
852 }
853 
854 static int ceph_remount(struct super_block *sb, int *flags, char *data)
855 {
856 	sync_filesystem(sb);
857 	return 0;
858 }
859 
860 static const struct super_operations ceph_super_ops = {
861 	.alloc_inode	= ceph_alloc_inode,
862 	.free_inode	= ceph_free_inode,
863 	.write_inode    = ceph_write_inode,
864 	.drop_inode	= generic_delete_inode,
865 	.evict_inode	= ceph_evict_inode,
866 	.sync_fs        = ceph_sync_fs,
867 	.put_super	= ceph_put_super,
868 	.remount_fs	= ceph_remount,
869 	.show_options   = ceph_show_options,
870 	.statfs		= ceph_statfs,
871 	.umount_begin   = ceph_umount_begin,
872 };
873 
874 /*
875  * Bootstrap mount by opening the root directory.  Note the mount
876  * @started time from caller, and time out if this takes too long.
877  */
878 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
879 				       const char *path,
880 				       unsigned long started)
881 {
882 	struct ceph_mds_client *mdsc = fsc->mdsc;
883 	struct ceph_mds_request *req = NULL;
884 	int err;
885 	struct dentry *root;
886 
887 	/* open dir */
888 	dout("open_root_inode opening '%s'\n", path);
889 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
890 	if (IS_ERR(req))
891 		return ERR_CAST(req);
892 	req->r_path1 = kstrdup(path, GFP_NOFS);
893 	if (!req->r_path1) {
894 		root = ERR_PTR(-ENOMEM);
895 		goto out;
896 	}
897 
898 	req->r_ino1.ino = CEPH_INO_ROOT;
899 	req->r_ino1.snap = CEPH_NOSNAP;
900 	req->r_started = started;
901 	req->r_timeout = fsc->client->options->mount_timeout;
902 	req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
903 	req->r_num_caps = 2;
904 	err = ceph_mdsc_do_request(mdsc, NULL, req);
905 	if (err == 0) {
906 		struct inode *inode = req->r_target_inode;
907 		req->r_target_inode = NULL;
908 		dout("open_root_inode success\n");
909 		root = d_make_root(inode);
910 		if (!root) {
911 			root = ERR_PTR(-ENOMEM);
912 			goto out;
913 		}
914 		dout("open_root_inode success, root dentry is %p\n", root);
915 	} else {
916 		root = ERR_PTR(err);
917 	}
918 out:
919 	ceph_mdsc_put_request(req);
920 	return root;
921 }
922 
923 
924 
925 
926 /*
927  * mount: join the ceph cluster, and open root directory.
928  */
929 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc)
930 {
931 	int err;
932 	unsigned long started = jiffies;  /* note the start time */
933 	struct dentry *root;
934 
935 	dout("mount start %p\n", fsc);
936 	mutex_lock(&fsc->client->mount_mutex);
937 
938 	if (!fsc->sb->s_root) {
939 		const char *path;
940 		err = __ceph_open_session(fsc->client, started);
941 		if (err < 0)
942 			goto out;
943 
944 		/* setup fscache */
945 		if (fsc->mount_options->flags & CEPH_MOUNT_OPT_FSCACHE) {
946 			err = ceph_fscache_register_fs(fsc);
947 			if (err < 0)
948 				goto out;
949 		}
950 
951 		if (!fsc->mount_options->server_path) {
952 			path = "";
953 			dout("mount opening path \\t\n");
954 		} else {
955 			path = fsc->mount_options->server_path + 1;
956 			dout("mount opening path %s\n", path);
957 		}
958 
959 		ceph_fs_debugfs_init(fsc);
960 
961 		root = open_root_dentry(fsc, path, started);
962 		if (IS_ERR(root)) {
963 			err = PTR_ERR(root);
964 			goto out;
965 		}
966 		fsc->sb->s_root = dget(root);
967 	} else {
968 		root = dget(fsc->sb->s_root);
969 	}
970 
971 	fsc->mount_state = CEPH_MOUNT_MOUNTED;
972 	dout("mount success\n");
973 	mutex_unlock(&fsc->client->mount_mutex);
974 	return root;
975 
976 out:
977 	mutex_unlock(&fsc->client->mount_mutex);
978 	return ERR_PTR(err);
979 }
980 
981 static int ceph_set_super(struct super_block *s, void *data)
982 {
983 	struct ceph_fs_client *fsc = data;
984 	int ret;
985 
986 	dout("set_super %p data %p\n", s, data);
987 
988 	s->s_flags = fsc->mount_options->sb_flags;
989 	s->s_maxbytes = MAX_LFS_FILESIZE;
990 
991 	s->s_xattr = ceph_xattr_handlers;
992 	s->s_fs_info = fsc;
993 	fsc->sb = s;
994 	fsc->max_file_size = 1ULL << 40; /* temp value until we get mdsmap */
995 
996 	s->s_op = &ceph_super_ops;
997 	s->s_d_op = &ceph_dentry_ops;
998 	s->s_export_op = &ceph_export_ops;
999 
1000 	s->s_time_gran = 1;
1001 	s->s_time_min = 0;
1002 	s->s_time_max = U32_MAX;
1003 
1004 	ret = set_anon_super(s, NULL);  /* what is that second arg for? */
1005 	if (ret != 0)
1006 		goto fail;
1007 
1008 	return ret;
1009 
1010 fail:
1011 	s->s_fs_info = NULL;
1012 	fsc->sb = NULL;
1013 	return ret;
1014 }
1015 
1016 /*
1017  * share superblock if same fs AND options
1018  */
1019 static int ceph_compare_super(struct super_block *sb, void *data)
1020 {
1021 	struct ceph_fs_client *new = data;
1022 	struct ceph_mount_options *fsopt = new->mount_options;
1023 	struct ceph_options *opt = new->client->options;
1024 	struct ceph_fs_client *other = ceph_sb_to_client(sb);
1025 
1026 	dout("ceph_compare_super %p\n", sb);
1027 
1028 	if (compare_mount_options(fsopt, opt, other)) {
1029 		dout("monitor(s)/mount options don't match\n");
1030 		return 0;
1031 	}
1032 	if ((opt->flags & CEPH_OPT_FSID) &&
1033 	    ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
1034 		dout("fsid doesn't match\n");
1035 		return 0;
1036 	}
1037 	if (fsopt->sb_flags != other->mount_options->sb_flags) {
1038 		dout("flags differ\n");
1039 		return 0;
1040 	}
1041 	return 1;
1042 }
1043 
1044 /*
1045  * construct our own bdi so we can control readahead, etc.
1046  */
1047 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1048 
1049 static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc)
1050 {
1051 	int err;
1052 
1053 	err = super_setup_bdi_name(sb, "ceph-%ld",
1054 				   atomic_long_inc_return(&bdi_seq));
1055 	if (err)
1056 		return err;
1057 
1058 	/* set ra_pages based on rasize mount option? */
1059 	sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT;
1060 
1061 	/* set io_pages based on max osd read size */
1062 	sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT;
1063 
1064 	return 0;
1065 }
1066 
1067 static struct dentry *ceph_mount(struct file_system_type *fs_type,
1068 		       int flags, const char *dev_name, void *data)
1069 {
1070 	struct super_block *sb;
1071 	struct ceph_fs_client *fsc;
1072 	struct dentry *res;
1073 	int err;
1074 	int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
1075 	struct ceph_mount_options *fsopt = NULL;
1076 	struct ceph_options *opt = NULL;
1077 
1078 	dout("ceph_mount\n");
1079 
1080 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1081 	flags |= SB_POSIXACL;
1082 #endif
1083 	err = parse_mount_options(&fsopt, &opt, flags, data, dev_name);
1084 	if (err < 0) {
1085 		res = ERR_PTR(err);
1086 		goto out_final;
1087 	}
1088 
1089 	/* create client (which we may/may not use) */
1090 	fsc = create_fs_client(fsopt, opt);
1091 	if (IS_ERR(fsc)) {
1092 		res = ERR_CAST(fsc);
1093 		goto out_final;
1094 	}
1095 
1096 	err = ceph_mdsc_init(fsc);
1097 	if (err < 0) {
1098 		res = ERR_PTR(err);
1099 		goto out;
1100 	}
1101 
1102 	if (ceph_test_opt(fsc->client, NOSHARE))
1103 		compare_super = NULL;
1104 	sb = sget(fs_type, compare_super, ceph_set_super, flags, fsc);
1105 	if (IS_ERR(sb)) {
1106 		res = ERR_CAST(sb);
1107 		goto out;
1108 	}
1109 
1110 	if (ceph_sb_to_client(sb) != fsc) {
1111 		destroy_fs_client(fsc);
1112 		fsc = ceph_sb_to_client(sb);
1113 		dout("get_sb got existing client %p\n", fsc);
1114 	} else {
1115 		dout("get_sb using new client %p\n", fsc);
1116 		err = ceph_setup_bdi(sb, fsc);
1117 		if (err < 0) {
1118 			res = ERR_PTR(err);
1119 			goto out_splat;
1120 		}
1121 	}
1122 
1123 	res = ceph_real_mount(fsc);
1124 	if (IS_ERR(res))
1125 		goto out_splat;
1126 	dout("root %p inode %p ino %llx.%llx\n", res,
1127 	     d_inode(res), ceph_vinop(d_inode(res)));
1128 	return res;
1129 
1130 out_splat:
1131 	ceph_mdsc_close_sessions(fsc->mdsc);
1132 	deactivate_locked_super(sb);
1133 	goto out_final;
1134 
1135 out:
1136 	destroy_fs_client(fsc);
1137 out_final:
1138 	dout("ceph_mount fail %ld\n", PTR_ERR(res));
1139 	return res;
1140 }
1141 
1142 static void ceph_kill_sb(struct super_block *s)
1143 {
1144 	struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1145 	dev_t dev = s->s_dev;
1146 
1147 	dout("kill_sb %p\n", s);
1148 
1149 	ceph_mdsc_pre_umount(fsc->mdsc);
1150 	flush_fs_workqueues(fsc);
1151 
1152 	generic_shutdown_super(s);
1153 
1154 	fsc->client->extra_mon_dispatch = NULL;
1155 	ceph_fs_debugfs_cleanup(fsc);
1156 
1157 	ceph_fscache_unregister_fs(fsc);
1158 
1159 	destroy_fs_client(fsc);
1160 	free_anon_bdev(dev);
1161 }
1162 
1163 static struct file_system_type ceph_fs_type = {
1164 	.owner		= THIS_MODULE,
1165 	.name		= "ceph",
1166 	.mount		= ceph_mount,
1167 	.kill_sb	= ceph_kill_sb,
1168 	.fs_flags	= FS_RENAME_DOES_D_MOVE,
1169 };
1170 MODULE_ALIAS_FS("ceph");
1171 
1172 int ceph_force_reconnect(struct super_block *sb)
1173 {
1174 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1175 	int err = 0;
1176 
1177 	ceph_umount_begin(sb);
1178 
1179 	/* Make sure all page caches get invalidated.
1180 	 * see remove_session_caps_cb() */
1181 	flush_workqueue(fsc->inode_wq);
1182 
1183 	/* In case that we were blacklisted. This also reset
1184 	 * all mon/osd connections */
1185 	ceph_reset_client_addr(fsc->client);
1186 
1187 	ceph_osdc_clear_abort_err(&fsc->client->osdc);
1188 
1189 	fsc->blacklisted = false;
1190 	fsc->mount_state = CEPH_MOUNT_MOUNTED;
1191 
1192 	if (sb->s_root) {
1193 		err = __ceph_do_getattr(d_inode(sb->s_root), NULL,
1194 					CEPH_STAT_CAP_INODE, true);
1195 	}
1196 	return err;
1197 }
1198 
1199 static int __init init_ceph(void)
1200 {
1201 	int ret = init_caches();
1202 	if (ret)
1203 		goto out;
1204 
1205 	ceph_flock_init();
1206 	ret = register_filesystem(&ceph_fs_type);
1207 	if (ret)
1208 		goto out_caches;
1209 
1210 	pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1211 
1212 	return 0;
1213 
1214 out_caches:
1215 	destroy_caches();
1216 out:
1217 	return ret;
1218 }
1219 
1220 static void __exit exit_ceph(void)
1221 {
1222 	dout("exit_ceph\n");
1223 	unregister_filesystem(&ceph_fs_type);
1224 	destroy_caches();
1225 }
1226 
1227 module_init(init_ceph);
1228 module_exit(exit_ceph);
1229 
1230 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1231 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1232 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1233 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1234 MODULE_LICENSE("GPL");
1235